|Publication number||US7779235 B2|
|Application number||US 11/671,627|
|Publication date||Aug 17, 2010|
|Filing date||Feb 6, 2007|
|Priority date||Feb 6, 2007|
|Also published as||CN101241429A, CN101241429B, US20080189520|
|Publication number||11671627, 671627, US 7779235 B2, US 7779235B2, US-B2-7779235, US7779235 B2, US7779235B2|
|Inventors||Deepak K. Singh, Francois Ibrahim Atallah|
|Original Assignee||International Business Machines Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (73), Non-Patent Citations (7), Referenced by (1), Classifications (6), Legal Events (2) |
|External Links: USPTO, USPTO Assignment, Espacenet|
Using performance data for instruction thread direction
US 7779235 B2
A method for dispatching instructions in the data processing system, having in memory for storing instructions and a plurality of central processing units, where each central processing unit includes a circuit to provide data indicating internal performance, the method having steps of receiving internal performance data signals from a pool of central processing units, selecting a central processing unit according to the received internal performance data and dispatching instructions from the memory to the selected central processing unit.
1. A method of dispatching instructions in a data processing system having a memory for storing instructions and a plurality of central processing units (CPUs), each CPU providing a signal indicating CPU frequency response data, the method comprising the steps of:
for each CPU, measuring a frequency from a frequency response circuit at the CPU, measuring a CPU temperature, using the measured temperature for accessing a previously stored predicted frequency, combining the predicted frequency and the measured frequency to provide a CPU frequency response and determining which of the plurality of CPUs has the highest frequency response,
selecting CPU with the highest frequency response, and
dispatching instructions from said memory to the selected CPU.
2. A method according to claim 1, wherein said selecting step includes the step of determining which of the plurality of CPUs is able to receive instructions for execution.
3. A method according to claim 1 wherein said step of measuring frequency includes receiving a signal from a ring oscillator contained at the frequency response circuit.
4. A data processing system comprising:
a plurality of central processing units (CPUs) where each CPU having an frequency response measurement circuit providing an frequency response data signal,
a memory including program instructions for execution by a CPU and connected to said plurality of CPUs,
a selection circuit connected to the plurality of CPUs, for each CPU, measuring a frequency from a frequency response circuit at the CPU, measuring a CPU temperature, using the measured temperature for accessing a previously stored predicted frequency, combining the predicted frequency and the measured frequency to provide a CPU frequency response and providing a signal indicating a selected CPU for the CPU having the highest frequency response, and
an instruction dispatching circuit connected to the memory and the plurality of CPUs and providing instructions to the selected CPU.
5. A data processing system according to claim 4 wherein said selection circuit includes a circuit to determine which of the plurality of CPUs are idle and is able to receive instructions for execution.
6. A computer memory storing a computer program product, the computer program product containing instructions for execution by a computer, which, when executed by the computer, cause the computer to implement a method for dispatching instructions in a data processing system having a memory for storing instructions and a plurality of central processing units (CPUs), each CPU providing a signal indicating CPU frequency response data, the method comprising the steps of:
for each CPU, measuring a frequency from a frequency response circuit at the CPU, measuring a CPU temperature, using the measured temperature for accessing a previously stored predicted frequency, combining the predicted frequency and the measured frequency to provide a CPU frequency response and determining which of the plurality of CPUs has the highest frequency response,
selecting CPU with the highest frequency response, and
dispatching instructions from said memory to the selected CPU.
7. The computer memory stored computer program product of claim 6, wherein said selecting step includes the step of determining which of the plurality of CPUs is able to receive instructions for execution.
8. The computer memory stored computer program product of claim 6 wherein said step of measuring frequency includes receiving a signal from a ring oscillator contained at the circuit.
9. A method according to claim 3 wherein said ring oscillator is connected to a bandgap voltage source.
10. A data processing system according to claim 5 wherein each frequency response measurement circuit includes a ring oscillator connected to a bandgap voltage source.
11. The computer memory stored computer program product of claim 8 wherein said ring oscillator is connected to a bandgap voltage source.
This application is related to the following co-pending U.S. Patent Applications filed on the same day as the present application and having the same assignee: “On-Chip Adaptive Voltage Compensation,” (U.S. patent application Ser. No. 11/671,485); “Using Temperature Data for Instruction Thread Direction,” (U.S. patent application Ser. No. 11/671,640); “Using IR Drop Data for Instruction Thread Direction,” (U.S. patent application Ser. No. 11/671,613); “Integrated Circuit Failure Prediction,” (U.S. patent application Ser. No. 11/671,599); “Instruction Dependent Dynamic Voltage Compensation,” (U.S. patent application Ser. No. 11/671,579); “Temperature Dependent Voltage Source Compensation,” (U.S. patent application Ser. No. 11/671,568); “Fan Speed Control from Adaptive Voltage Supply,” (U.S. patent application Ser. No. 11/671,555); and “Digital Adaptive Voltage Supply,” (U.S. patent application Ser. No. 11/671,531); each assigned to the IBM Corporation and herein incorporated by reference.
BACKGROUND OF THE INVENTION
1. Technical Field
The present invention relates in general to a system and method for instruction thread distribution. In particular, the present invention relates to a system and method for directing instruction thread distribution according to the performance data of the circuitry to execute the instructions.
2. Description of the Related Art
Many modern data processing systems include multiple central processing unit cores (CPUs) in the system. These data processing systems will execute instructions of a single program across these multiple central processing unit cores. The single program includes many instructions to be executed in the central processing units. One technique to employ the multiple central processing unit cores in the execution of these instructions is to divide the instructions into groups of instructions or threads. Then each thread is directed to a central processing unit for execution. Several prior art patents address the use of instruction threads in a processor and the control of execution of these instruction threads. These patents include U.S. Pat. No. 7,093,109 entitled “Network Processor which makes Thread Execution Control Decisions Based on Latency Event Lengths”; U.S. Pat. No. 6,076,157 entitled “Method and Apparatus to Force a Thread Switch in a Multithreaded Processor”; U.S. Pat. No. 6,212,544 entitled “Altering Thread Priorities in a Multithreaded Processor”; and U.S. Pat. No. 6,625,637 entitled “Deterministic and Preemptive Thread Scheduling and Its Use in the Debugging Multithreaded Applications.”
In a multiple central processing unit data processing system, it is helpful to know the physical conditions of the central processing unit cores that will be receiving the instruction threads. To obtain the maximum performance within the data processing system, distribution of the instruction threads for execution should be made to the central processing units that are able to execute these instruction threads efficiently. One physical condition of the central processing unit cores is the performance data or frequency response that is measured in terms of clock frequency and it is inherently due to the manufacturing process. The number of CPU cores that can be implemented on a chip is proportional to the area of the chip. But as the chip area increases, each separation between CPU cores located near the opposite edges of the chip also increases. In a chip with large area, the performance of individual devices contained in cores that are not within close spatial proximity differs due to minor changes in semiconductor manufacturing process seen by distant cores. The net effect of this is that cores that are separated offer different frequency response or performance. The higher the performance data for the core, the more efficient the central processing unit will be.
In accordance with the present invention, a method for dispatching instructions in a data processing system having multiple central processing units where each central processing unit provides performance data, the method including the steps of receiving the performance data from the central processing units, selecting a central processing unit, according to the received performance data, and dispatching instructions from memory to the selected central processing unit.
In one embodiment of the present invention, a data processing system is provided, that includes several central processing unit cores, a memory including program instructions for execution by a central processing unit, a selection circuit connected to receive performance data provided by the central processing units, and selecting a central processing unit according to the performance data received, and an instruction dispatch a circuit connected to the memory and the central processing units to the selected central processing unit.
BRIEF DESCRIPTION OF THE DRAWINGS
The present invention may be better understood, and its numerous objects, features, and advantages made apparent to those skilled in the art by referencing the accompanying drawings.
FIG. 1 is a schematic diagram of a simple embodiment of the temperature measurement circuit;
FIG. 2 is a schematic diagram of a second embodiment of the temperature measurement circuit;
FIG. 3 is a schematic diagram of the two ring oscillator circuit that provides input for the frequency response measurement and provides the IR drop measurement;
FIG. 4 is a schematic diagram of the preferred embodiment of the adaptive voltage compensation circuit;
FIG. 5 is a flow chart representing the operation of the adaptive voltage compensation circuit;
FIG. 6 is a block diagram of a digital implementation of the adaptive voltage compensation circuit;
FIG. 7 is a second and more detailed block diagram of the implementation of FIG. 6;
FIG. 8 is an illustration of the location of adaptive voltage compensation circuits on multiple cores;
FIG. 9 is a flow diagram illustrating performance data being obtained and instruction threads being directed accordingly; and
FIG. 10 is a diagram illustrating the instruction distribution process, connected to the program memory, and the multiple central processing units.
The following is intended to provide a detailed description of an example of the invention and should not be taken to be limiting of the invention itself. Rather, any number of variations may fall within the scope of the invention, which is defined in the claims following the description.
The present invention provides a system to measure temperature within a single central processing unit. This is actually accomplished by providing an adaptive power supply (APS) for each central processing unit. Each of these adaptive power supplies determines operating conditions on an integrated circuit and adjust voltage (Vdd) provided to the integrated circuit to either increase performance of the integrated circuit or save power expended by the integrated circuit.
In a preferred embodiment of these adaptive power supplies, three physical condition measurements are made. The first is temperature, which is measured by a thermal diode on the surface of the integrated circuit. The second is the IR (voltage) drop measured by two ring oscillator circuits and the third is the frequency performance of the integrated circuit measured by a single loop oscillator compared to stored predetermined performance values.
The complete control signal provided to the voltage regulation circuit is:
Total Vdd scaling=Frequency response scaling+Temperature related Vdd scaling+IR drop related scaling
All of the measurement circuits are contained on the surface of this integrated circuit device in the preferred embodiment. These measurements are then used to scale an input control signal to a voltage regulation circuit also contained on the surface of the integrated circuit device or alternatively on another integrated circuit. The output of this voltage regulation device provides the integrated circuit operating voltage (Vdd). Thus the voltage supplied to the integrated circuit can be adjusted to either save power or increase performance dynamically during the operation of the chip by under program control. Further the integrated circuit voltage and, therefore, performance can be changed in anticipation of operating environment changes such as a sleep state or the execution of instructions requiring high circuit performance.
This is a dynamic method of varying voltage that takes into account the specifics of the semiconductor manufacturing process, temperature and IR drop effects simultaneously. This method uses available on-chip data to compute adjustment in voltage necessary to either meet target performance or decrease power consumption. The two goals are met using the same circuit. Another advantage of using this method is the flexibility it offers to the users in terms of programmability. On chip voltage can be artificially varied by writing into special registers which provide values used by the power management circuitry to provide the supply voltage Vdd. This feature can be helpful when expecting instructions that require high circuit performance, essentially providing an “on-Demand” performance capability. In other words, to provide on request, additional circuit supply voltage to increase circuit performance.
This method is not limited to a specific technology or type of circuit. It can be applied to a broad type of integrated circuits, especially those that need to deliver higher performance at lower power consumption.
This method also offers reduction in test time for identifying yield and voltage per module. It is a dynamic solution unlike previous static solutions (fuses, etc) that takes into account effects of IR drop.
FIG. 1 is a schematic diagram of one embodiment of the thermal measurement circuit 125 shown connected to the voltage regulation circuit which provides the integrated circuit voltage source (Chip Vdd). This measurement circuit includes a current source 100 connected to the voltage source. This current source 100 is also connected by a line 103 to a thermal diode 102 also connected to ground. The voltage across the thermal diode 102 indicates the measured temperature of this integrated circuit. This thermal voltage signal is provided over line 103 to an analog comparator 106. The output of the comparator 106 is connected to an address counter 110 providing an address to a digital to analog (D to A) converter 114. The operating range for a thermal diode is commonly zero to 125° C. The address counter 110 includes a look up table with 128 entries. These entries correspond to 0 to 127 degrees C. Initially, the address counter 110 starts at zero degrees and increments upward each clock cycle. Each address is provided to the D to A converter 114 over line 112. In operation, the analog comparator 106 compares the output of the D to A converter 114 with the measured thermal voltage provided by the thermal diode 102. When the address counter 110 provides an output representing the same temperature as the thermal diode 102, the output voltage from the D to A converter 110 will be the same voltage as that provided by the thermal diode 102. The output of the analog comparator 106 will then be zero. The address counter 110 will then stop incrementing and provide a signal over line 116 to a delay lookup table (LUT) circuit 118. This value on line 116 is a digital signal representing the temperature measured by the thermal diode 102. This thermal voltage value is used to address a corresponding delay value in the delay lookup table circuit 118. The delay lookup table in circuit 118 is a table of pulse width values computed by a simulation of the performance of the integrated circuit. Each value represents the expected delay value computed for the temperature range of 0 to 127 degrees C. for expected integrated circuit performance.
To measure the process on the substrate, a ring oscillator connected to a temperature compensated voltage source (ex: a bandgap reference) is used. In this case, for a given temperature, the pulse width produced by the ring oscillator is a function of the process on the substrate since temperature and voltage are constant. By using a bandgap reference, the voltage applied to a ring oscillator can be kept constant. But the temperature of the substrate depends upon internal and external operating conditions and it cannot be held constant. To eliminate the effects of varying temperature, another scheme is used in this invention.
First, a target predicted circuit performance number (pcpn) is chosen. This number represents the expected circuit performance based on expected semiconductor manufacturing process. This number represents circuit performances expected under nominal applied voltage across the entire operating temperature range. For this pcpn, a simulation of the ring oscillator supplied by a constant voltage from a bandgap reference is carried out for the entire operating temperature range. This simulation yields pulse widths that are generated at a fixed voltage and pcpn values where only the temperature is varied across the entire operating temperature range. If the substrate pcpn is identical to the desired target performance, then the substrate would also yield identical pulse widths for each value of the operating temperature range.
If the substrate pcpn is different than the desired target performance, then the pulse widths produced by the substrate will be either shorter or longer than those produced by simulation depending upon whether the substrate pcpn was faster or slower than the desired target performance. So a comparison has to be made between the pulse width generated by the ring oscillator on the substrate with a simulated value of the pulse with at the value of the substrate temperature at a fixed voltage. The expected pulse width values at the desired target process for each temperature value within the desired operating temperature range are stored in a Look Up Table (LUT) (for example, 118 in FIG. 1) that is addressed by the current substrate temperature, i.e. based on the substrate temperature, the address pointer points to an entry in the LUT that contains the expected pulse width from the ring oscillator circuit at the desired process corner at a fixed bandgap voltage. For this invention, the operating temperature range is 0° C. to 127° C. and this range is divided into 128 steps of 1° C. each. This requires 128 entries in the LUT, one entry corresponding to each 1° C. rise in temperature.
This resulting pulse width value from the delay lookup table circuit 118 provides a voltage scaling signal in digital form which is converted to an analog voltage signal by D to A converter 122. This scaling voltage signal is provided to a voltage regulator 130 over line 124. The operation result of the circuit 125 would be to increase or decrease the resulting voltage of regulator circuit 130 (chip Vdd) based upon the measured temperature of the integrated circuit measured by thermal diode 102.
FIG. 2 is a second embodiment of the thermal measurement circuit illustrated in FIG. 1. The temperature measurement circuit 225 of FIG. 2 includes two current sources 200 and 202 which are selectively connected to a thermal diode 208 through a switch 204 connected by line 206. The diode is actually made up of a lateral PNP device fabricated in CMOS technology. The collector and base of this device are shorted leaving the diode between base and emitter.
Digital temperature sensors are based on the principle that the base-emitter voltage, VBE, of a diode-connected transistor is inversely proportional to its temperature. When operated over temperature, VBE exhibits a negative temperature coefficient of approximately −2 mV/° C. In practice, the absolute value of VBE varies from transistor to transistor. To nullify this variation, the circuit would have to calibrate each individual transistor. A common solution to this problem is to compare the change in VBE of the transistor when two different current values are applied to the emitter of the transistor.
Temperature measurements are made using a diode that is fed by 2 current sources, one at a time. Typically the ratio of these current sources is 10:1. The temperature measurement requires measuring the difference in voltage across the diode produced by applying two current sources.
Line 206 is connected to a “sample and hold” circuit 209 to sample and hold a voltage output of the thermal diode 208. The address counter circuit 222 operates identically to the address counter, circuit 110 of FIG. 1 previously discussed. Address counter circuit 222 increments an address every clock cycle which provides a digital signal representing the temperature range of zero to 127° C. over line 220 to the D to A converter 218 which converts this digital signal representing temperature to a voltage. This voltage signal is provided on line 215 to a second sample and hold circuit 213. Both the sample of the hold circuits 209 and 213 will sample and hold their respective voltages for the comparator 212 so that continuing small variations in temperature from the thermal diode 208 will not adversely affect the operation of this temperature measurement circuit 225. Upon reaching the measured temperature, the comparator 212 will provide a zero output over line 216 to the address counter 222 which provides a digital signal representing the measured temperature on line 224 to the delay lookup table circuit 226. The operation of the delay lookup table circuit 226 providing a digital delay value on line 228 to the D to A converter 230 is the same as previously discussed for the measurement circuitry 125 in FIG. 1.
FIG. 3 is a schematic diagram of the IR drop measurement circuit 325 which provides voltage scaling signal to a voltage regulator circuit 326. A band gap voltage source 300 is connected to a ring oscillator circuit 304. The ring oscillator circuit 304 consists of an odd number of inverters 302 connected in a loop or ring. The band gap source is obtained from the physical integrated circuit itself and is nominally 1.23 V. A second ring oscillator circuit 306 connected to the chip voltage source provides an output on line 314. The band gap ring oscillator provides an output on line 312. A phase detector 308 is connected to lines 312 and 314 to determine the difference or delay between the pulses provided by the two ring oscillator circuits 304 and 306. The phase detector 308 provides a voltage magnitude output and a voltage polarity output on lines 316 and 318 respectively which in combination represent the delay difference between the ring oscillator circuits 304 and 306. Lines 316 and 318 are input to a comparator 310 which provides a voltage scaling signal on line 322 to the voltage regulator 326. It should be understood that this voltage scaling signal on line 322 is based solely upon the IR drop of the integrated circuit. Based on the voltage scaling signal of line 322, voltage regulator 326 provides the appropriate chip Vdd value. In the preferred embodiment, the two ring oscillator circuits 304 and 306 should be located in close proximity to each other so that the effects of any irregularities across the surface of the integrated circuit will be minimized.
The frequency response of the integrated circuit (or performance of the integrated circuit) can be measured by using the output of a band gap voltage connected ring oscillator 304 on line 305 of FIG. 3 and the lookup table containing known delay values based on chip temperature from circuit 226 or FIG. 2. This is illustrated in combination with the IR drop measurement of circuit 325 and the temperature measurement of circuit 225 in FIG. 4. In the IR drop measurement circuit 325, the band gap connected ring oscillator 304 provides a second signal connected to an integrator circuit 414, which takes the pulse signal from the band gap connected ring oscillator 304 of circuit 325 and converts it into a voltage which is then provided to difference circuit 416. Another input line 415 to the difference circuit 416 is compared to the delay voltage signal output from the D to A converter 230 representing the expected delay based on the measured temperature. The output of this difference circuit 416 represents a voltage indicative of the integrated circuit frequency response or performance of the integrated circuit. More specifically, this signal provided to multiplexer 418 represents the actual integrated circuit performance compared to the expected integrated circuit performance for that temperature. If the expected delay signal on line 415 is less than the delay signal from integrator circuit 414, the chip is performing below expectations and the voltage Vdd should be increased. Conversely, if the expected delay on line 415 is greater than the delay signal from integrator circuit 414, the chip is performing above expectations and the voltage Vdd could be lowered to save power.
FIG. 4 also illustrates the preferred embodiment of the invention combining the temperature measurement circuit 325 output, the IR drop measurement circuit 325 output with the frequency response measurement as discussed above. In this embodiment, the temperature measurement circuit includes a lookup table address register 400 connected to the address counter 210 by line 402 to provide an initial address or to provide an artificially changed temperature that would result in an artificially changed voltage scaling signal. Also, the lookup table data register 406 is provided that may provide a directed input into the delay lookup table 226 shown in block 450 where block 450 also contains other circuit elements for frequency response measurement. This can be used to provide entries into the delay lookup table or provide bypass data output directly to multiplexer 410 which is input to the D to A converter 230. In this manner, a programmer could directly control the delay value, which is used to compute the voltage scaling signal on line 428. The output of the D to A converter 230 is provided on line 415 directly to the difference circuit 416 and to the multiplexer 418. In this manner the multiplexer 418 may bypass the difference circuit 416 and only provide the temperature dependant table delay value to the driver 420. The driver 420 is connected to a register 408 by line 438 which can be used to control the amount of signal output on line 424 to the summing circuit 426. Likewise, in circuit 325, register 432 provides on line 434, a signal that can be used to vary the amount of the scaling signal output from the circuit 325 to the summing circuit 426. The output from summing circuit 426 is the voltage scaling signal on line 428 and is provided to the voltage regulator 436 which in turn provides the integrated circuit voltage (chip Vdd) 440.
FIG. 5 is a process flow chart representing the operation of the invention. It is important understand, that FIG. 5 is not a flow chart representing software execution but of a simultaneous process producing the voltage scaling signal previously discussed in the operation of the different functional units of the present invention. The discussion of this flowchart of FIG. 5 will also reference FIGS. 2, 3 and 4 respectively. In the start phase 500, path 524 illustrates the simultaneous operation of the different aspects of this invention. In step 502, the thermal diode 208 provides an output voltage indicating the measured circuit temperature on line 506 to process block 504. Process block 504 represents the operation of the address counter 222, the D to A converter 218 and the voltage comparator 212 (of FIG. 2) in determining a digital signal representative of the circuit temperature as previously discussed. Referring to FIG. 5, this digital temperature is provided on path 530 to the delay lookup table in step 506 which provides a digital signal representative of the delay on path 534 to the D to A conversion step 508 resulting in the delay signal voltage provided to the comparator 514 over path 536.
Returning to path 524, the frequency response value measured in block 510 is provided in path 528 to both the integration block 512 and to the compare block 520 by line 538 as discussed in FIG. 4. The integration circuit 414 of FIG. 4 provides the frequency response measurement signal to the compare block 514 over path 542 which is then compared to the delay signal on path 536. This result of this comparison is provided on path 544. Returning to path 524, the measurement of the IR drop from the ring oscillator 306 connected to the chip voltage supply is compared with the ring oscillator 304 connected to the band gap voltage source in step 520. The output on path 540 represents the IR drop portion of the voltage scaling signal and is combined in step 516 to produce the overall voltage scaling signal 546 provided to the regulator 436 in step 522. It is important understand that this voltage scaling signal results from the combination of the measurements for temperature, IR drop and circuit frequency response.
Digital Implementation of the Adaptive Voltage Supply
FIG. 6 is a block diagram of an embodiment of the digital adaptive voltage supply. Block 604 represents the temperature sensor previously discussed in FIGS. 1, 2 and 4. Register 600 provides an address into the temperature sensor tables, as previously discussed. The output of the temperature sensor block 604 on line 606 is provided to the pulse width table 608. This table 608 is also connected by line 622 to a data register 610. The data register 610 provides the ability to input a value into either the pulse width table 608 or to the multiplexer 612. In this manner, the adaptive power management unit 622 may provide inputs into data register 610 which is substituted by a multiplexer 612 for a pulse width value. In other words, a computer program providing control of the operation of the adaptive power management unit 622 can directly control the value in the data register 610 and thus indirectly control the voltage scaling computation from this point in the block diagram.
The bandgap reference circuit 618 and the Vdd reference circuit 632 are similar to those discussed and illustrated as block 325 in FIG. 4. However, the output of the bandgap reference circuit 618 and chip Vdd reference circuit 632 are combined in a difference circuit 642 that provides an output on line 640. The bandgap reference circuit 618 also provides an output that is combined with the output from the multiplexer 612 in the difference circuit 665. This difference circuit 665 provides an output on line 667.
One distinction from the adaptive voltage supply illustrated in FIG. 4 is the inclusion of the process sensor registers 676 connected to line 667 and the IR drop register 647 connected to line 640. Since the data on lines 667 and 640 are digital, these registers 676 and 647 may receive the values on these lines respectively. Alternatively, register 676 can receive an input on line 680 as can register 647 receive an input on line 637. In other words, both these registers are read/write registers. Returning to line 667, its value is input to a multiplier circuit 671 which receives an input from register 668 that provides a weighting value. In this embodiment, a weighting value can be used to increase or decrease the influence of the process number that results from either the difference circuit 665 or the process sensor register 678. Registers 668 receives an input on line 678 from the adaptive power management unit 622. The result of the multiplier circuit 671 is provided to the adding circuit 654. Line 640 also provides an input to a multiplier circuit 635 which receives a weighting value from the IR drop weight register 636. Like the process weight register 668, the IR drop weight register 636 receives an input on line 684 from the adaptive power management unit 622. The output of multiplier 635 is provided to the summing circuit 654 on line 652. The output from the summing circuit 654 is provided on line 650 to another multiplier 657, which is connected to a regulator weight register 660. This register, connected by line 682 to the adaptive power management unit allows program control of output of the scaling signal of the power supply itself. Therefore, by providing a weighting value in the register 660, the output on line 662 of the overall scaling circuitry can be regulated. Also in FIG. 6, there is a power supervisor circuit 627 which represents the interface to the computing system that permits for overall will control over this digital adaptive voltage supply through line 629 to the adaptive power management unit 622. The registers 600, 610, 676, 668, 660, 636, and 647 are read/write registers. Thus, the power supervisor 627 through the adaptive power management unit 622 can exercise total monitoring and regulation over the operation of the digital adaptive voltage supply.
FIG. 7 is a more detailed diagram of the block diagram of FIG. 6 further showing the Process Vt shift. As the part ages, the Vt for its devices shifts resulting in slower performance. This register 712 connected to the differencing circuit 718 which stores. The Process Vt shift register 712 stores the pulse width value generated by the ring oscillator 744. As the part ages, for the same value of temperature and at bandgap voltage, the value written into this register will become larger indicating that the part is slowing down. By periodically comparing the value stored in this register with a pre-calculated pulse width value (estimated at 80% of the final pulse width achieved at End of Life for the part) for a given temperature, it can be determined when the part has reached the 80% point of its End of Life Vt shift and a signal will be generated that this part may need to be replaced soon. In one embodiment, this register 712 is a read-only register where the value is written into the register based upon user control (i.e. a user can decide when the ring oscillator 744 pulse width data can be written into this register 712, but the user cannot write or overwrite the value of this register 712).
For thread re-direction, this register is not really used, but it is described here for the sake of completeness.
FIG. 8 is a diagram illustrated embodiment, where multiple CPU cores are located on a single semiconductor substrate 800. Each of the cores 802, 804, 806 and 808 are identical in this illustrated embodiment. However it should be apparent that the functionality of the cores is not relevant to the application of this invention as long as individual adaptive voltage supplies are located in each of the cores. In FIG. 8, the view of CPU core 804 is exploded into a view 810 that includes the CPU itself plus, on the surface of this core, an adaptive voltage supply 812 connected by a line 815 to a power supervisor 817. In operation, the power supervisor 817 represents the programmable control over all of the adaptive voltage supplies on all of the cores in the system. By using the registers discussed in FIGS. 6 and 7, the power supervisor 817 can control and monitor the operation of each adaptive voltage supply.
FIG. 9 is a flowchart implemented as a computer program product illustrating the operation of the power supervisor in controlling the adaptive voltage supply. The thermal diode voltage is read in step 900 which is connected to the process 922 that determines the temperature for the lookup table in step 925 to determine the measured temperature value which is provided to the differencing block 932 by line 979 which is also connected to the measured value register 980. Simultaneously, the first process sensing ring oscillator is read in block 928. This frequency with value is provided on line 932 to the write process shift register 926 and a difference circuit 932. Also simultaneously, the second process sensor ring oscillator circuit is read in block 940. Its output is provided of line 942 to the difference circuit 944 where the difference between the first and second ring oscillator circuits is provided on line 946.
FIG. 9 illustrates software control over the adaptive voltage supply previously discussed. Block 950 initiates the software or override capability through decision 954 from line 952. If a software or override is to take place, then the input measured IR drop value in block 962 would not be provided, but rather a software input value in block 960 would be provided over line 964 to the IR drop register 966.
In a similar manner, block 902 controls the process value that is used by the adaptive voltage supply. When a software control is implemented, a signal is provided on line 904 to the decision block 906. If an override by a software input is to take place, then the software input value in block 912 is provided by line 916 to the write process register 918 instead of the measured process of block 914. As shown, the inputted measured process value in block 914 is received via line 934 from the difference circuit 932 at this point. The software controls both the write process register in block 918 and the write IR drop register in block 966. Both the IR drop data and the process data are summed in block 936 to provide the overall voltage scaling signal that is output to the voltage regulator at 938 to provide the Vdd supply voltage to the integrated circuit.
Also in a similar manner, block 970 provides a user or software override in order to provide a substituted temperature value in place of the measured temperature value. This is done by providing a signal on line 974 to a decision process 972 if the software is to override the measured value, a signal is sent online 978, to access the software provided temperature value in block 982, which is written by line 984 into the write temperature register 986. However if there is no software override, the decision block 972 provides a signal on line 976 to the register 980 which receives the temperature from line 924 as previously discussed.
It should also be apparent to those skilled in the art that the use of weight registers also provides a greater degree of software control over the operation of the adaptive voltage supply. Therefore by accessing these registers, the power supervisor can both monitor and regulate the operation of each of the adaptive voltage supplies that are contained on the integrated circuit.
When multiple central processing unit cores are contained in the data processing system and each central processing unit core includes its own adaptive power supply, circuitry can be provided to collect data from each of the adaptive power supplies and provide it to supervisory software to enhance the efficient operation of the data processing system. One example is the use of temperature data obtained from the adaptive power supplies distributed on the central processing units. It has been determined that the cooler central processing units will execute instructions more efficiently. The temperature information can be used to select a central processing unit for executing a group of instructions or instructions in a thread. This is illustrated in FIG. 9 by process 990, which is connected by line 988 to the right temperature register 986 shown. In fact, process 990 is connected to all of the adaptive powers supplies to collect this temperature data from each of the adaptive power supplies. Once collected the process determines the central processing units with the lowest temperature. Then in process 992 a determination is made of which central processing unit cores are available for executing instructions. This process 982 then selects the central processing unit that is the lowest temperature and is available. Then in process 984, the instruction thread or group of instructions is dispatched to the selected central processing unit for execution.
While this discussed embodiment shows only a single voltage control circuit on the integrated circuit, it should be apparent that multiple voltage control circuits may be utilized to provide different voltages to different portions of the integrated circuit. This is illustrated in FIG. 10 which is similar to FIG. 8, and illustrating multiple central processing unit course on a single semiconductor substrate 800. In FIG. 10, CPU, 810 is the exploded view of the CPU 804 contained on the semiconductor substrate 800. CPU 810 includes an adaptive power supply, 812, that is connected by line 816 to an instruction dispatch process 818. The temperature data would be provided on line 816 to the instruction dispatch process which examines not only the temperature of each of the central processing units but also examines whether or not the central processing units are available to receive new instructions for execution. Upon determining the available central processing units available, then the central processing unit with the lowest temperature is selected to receive instructions for execution. These are instructions are obtained from the memory 824 over line 822, and they are provided to the central processing unit 810 by line 826. In this manner, instructions will be dispatched to the available central processing unit that can most efficiently execute them.
While particular embodiments of the present invention have been shown and described, it will be obvious to those skilled in the art that, based upon the teachings herein, that changes and modifications may be made without departing from this invention and its broader aspects. Therefore, the appended claims are to encompass within their scope all such changes and modifications as are within the true spirit and scope of this invention. Furthermore, it is to be understood that the invention is solely defined by the appended claims. It will be understood by those with skill in the art that if a specific number of an introduced claim element is intended, such intent will be explicitly recited in the claim, and in the absence of such recitation no such limitation is present. For non-limiting example, as an aid to understanding, the following appended claims contain usage of the introductory phrases “at least one” and “one or more” to introduce claim elements. However, the use of such phrases should not be construed to imply that the introduction of a claim element by the indefinite articles “a” or “an” limits any particular claim containing such introduced claim element to inventions containing only one such element, even when the same claim includes the introductory phrases “one or more” or “at least one” and indefinite articles such as “a” or “an”; the same holds true for the use in the claims of definite articles.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4417470||Sep 30, 1981||Nov 29, 1983||Otis Engineering Corporation||Electronic temperature sensor|
|US5375146 *||May 6, 1993||Dec 20, 1994||Comsat Corporation||Digital frequency conversion and tuning scheme for microwave radio receivers and transmitters|
|US5451894||Feb 24, 1993||Sep 19, 1995||Advanced Micro Devices, Inc.||Digital full range rotating phase shifter|
|US5457719||Aug 11, 1993||Oct 10, 1995||Advanced Micro Devices Inc.||All digital on-the-fly time delay calibrator|
|US5737342||May 31, 1996||Apr 7, 1998||Quantum Corporation||Method for in-chip testing of digital circuits of a synchronously sampled data detection channel|
|US5844826||Oct 18, 1996||Dec 1, 1998||Samsung Electronics Co., Ltd.||Leading zero count circuit|
|US5852616||Nov 17, 1997||Dec 22, 1998||Advanced Micro Devices, Inc.||On-chip operating condition recorder|
|US5990725||Jun 30, 1997||Nov 23, 1999||Maxim Integrated Products, Inc.||Temperature measurement with interleaved bi-level current on a diode and bi-level current source therefor|
|US6037732||Apr 21, 1998||Mar 14, 2000||Telcom Semiconductor, Inc.||Intelligent power management for a variable speed fan|
|US6047248||Oct 19, 1998||Apr 4, 2000||International Business Machines Corporation||Performance-temperature optimization by cooperatively varying the voltage and frequency of a circuit|
|US6058502||Sep 1, 1998||May 2, 2000||Nec Corporation||Diagnostic system analyzing frequency spectrum of electric power for diagnosing integrated circuit, method and information storage medium storing computer program for the method|
|US6070074||Apr 24, 1998||May 30, 2000||Trw Inc.||Method for enhancing the performance of a regenerative satellite communications system|
|US6076157||Oct 23, 1997||Jun 13, 2000||International Business Machines Corporation||Method and apparatus to force a thread switch in a multithreaded processor|
|US6111414||Jul 31, 1998||Aug 29, 2000||Georgia Tech Research Corporation||System, circuit, and method for testing an interconnect in a multi-chip substrate|
|US6125334||Apr 24, 1998||Sep 26, 2000||Texas Instruments Incorporated||Module-configurable full-chip power profiler|
|US6141762 *||Aug 3, 1998||Oct 31, 2000||Nicol; Christopher J.||Power reduction in a multiprocessor digital signal processor based on processor load|
|US6172611||Oct 19, 1998||Jan 9, 2001||Telcom Semiconductor, Inc.||Independent hardware thermal sensing and monitoring|
|US6212544||Oct 23, 1997||Apr 3, 2001||International Business Machines Corporation||Altering thread priorities in a multithreaded processor|
|US6351601||Feb 24, 1999||Feb 26, 2002||Micrel Incorporated||Fan speed control system having an integrated circuit fan controller and a method of using the same|
|US6479629||Jun 18, 2001||Nov 12, 2002||Pioneer Hi-Bred International, Inc.||Polypeptide; for use in the generation of disease resistant plants|
|US6481974||Feb 15, 2001||Nov 19, 2002||Sunonwealth Electric Machine Industry Co., Ltd.||Fan motor with constant speed control by a microprocessor system|
|US6591210||Nov 21, 2000||Jul 8, 2003||National Semiconductor Corporation||Circuit and method to combined trim and set point|
|US6625635||Nov 2, 1998||Sep 23, 2003||International Business Machines Corporation||Deterministic and preemptive thread scheduling and its use in debugging multithreaded applications|
|US6713996||Aug 28, 2002||Mar 30, 2004||Micron Technology, Inc.||Ultra low power tracked low voltage reference source|
|US6721581||May 5, 2000||Apr 13, 2004||Infineon Technologies Ag||Reprogrammable digital wireless communication device and method of operating same|
|US6721892||May 9, 2000||Apr 13, 2004||Palmone, Inc.||Dynamic performance adjustment of computation means|
|US6838917||Oct 7, 2002||Jan 4, 2005||Infineon Technologies Ag||Circuit configuration for processing data, and method for identifying an operating state|
|US6859113||Aug 29, 2003||Feb 22, 2005||Eta Sa Fabriques D'ebauches||Temperature compensation mechanism for a micromechanical ring resonator|
|US6897673||Mar 19, 2003||May 24, 2005||Lsi Logic Corporation||Method and integrated circuit for capacitor measurement with digital readout|
|US7086058||Jun 6, 2002||Aug 1, 2006||International Business Machines Corporation||Method and apparatus to eliminate processor core hot spots|
|US7093109||Apr 4, 2000||Aug 15, 2006||International Business Machines Corporation||Network processor which makes thread execution control decisions based on latency event lengths|
|US7096140||Oct 27, 2004||Aug 22, 2006||Kabushiki Kaisha Toshiba||Test system, test method and test program for an integrated circuit by IDDQ testing|
|US7100061||Jan 18, 2000||Aug 29, 2006||Transmeta Corporation||Adaptive power control|
|US7174194 *||Aug 17, 2001||Feb 6, 2007||Texas Instruments Incorporated||Temperature field controlled scheduling for processing systems|
|US7184936||Jul 12, 2004||Feb 27, 2007||Cisco Technology, Inc.||Timing variation measurement system and method|
|US7211977||Jul 13, 2004||May 1, 2007||Hewlett-Packard Development Company, L.P.||Pulse width modulation fan control|
|US7228446||Dec 21, 2004||Jun 5, 2007||Packet Digital||Method and apparatus for on-demand power management|
|US7282966 *||Sep 28, 2004||Oct 16, 2007||Intel Corporation||Frequency management apparatus, systems, and methods|
|US7307439||Aug 26, 2004||Dec 11, 2007||Nec Corporation||Semiconductor integrated circuit apparatus, measurement result management system, and management server|
|US7330081||Jul 7, 2005||Feb 12, 2008||Marvell Semiconductor Israel Ltd.||Digitally controlled oscillator and associated method|
|US7330983||Jun 14, 2004||Feb 12, 2008||Intel Corporation||Temperature-aware steering mechanism|
|US7437581||Sep 28, 2004||Oct 14, 2008||Intel Corporation||Method and apparatus for varying energy per instruction according to the amount of available parallelism|
|US20020046399||Oct 16, 2001||Apr 18, 2002||Anthony Debling||Interface device|
|US20020065049||Aug 17, 2001||May 30, 2002||Gerard Chauvel||Temperature field controlled scheduling for processing systems|
|US20030030483||Aug 12, 2002||Feb 13, 2003||Takahiro Seki||Semiconductor device|
|US20030057986||Sep 27, 2001||Mar 27, 2003||Brian Amick||Integrated on-chip process, temperature, and voltage sensor module|
|US20030067334||Oct 7, 2002||Apr 10, 2003||Eckhard Brass||Circuit configuration for processing data, and method for identifying an operating state|
|US20030079150||Oct 22, 2001||Apr 24, 2003||International Business Machines Corporation||Method and apparatus for reducing power comsumption of a processing integrated circuit|
|US20030126476||Jan 2, 2002||Jul 3, 2003||Greene Michael A.||Instruction scheduling based on power estimation|
|US20030184399 *||Jan 17, 2003||Oct 2, 2003||Cmc Electronique Inc./Cmc Electronics Inc.||Indirect temperature compensation process in an oscillator|
|US20040023688||Aug 5, 2002||Feb 5, 2004||Seyfollah Bazarjani||On-chip detection circuit off-chip mechanical switch's open and close actions|
|US20040025061||Oct 25, 2001||Feb 5, 2004||Lawrence Richard H.||Method and system for power reduction|
|US20040090216||Nov 7, 2002||May 13, 2004||International Business Machines Corporation||Method and apparatus for control of voltage regulation|
|US20040183613||Mar 21, 2003||Sep 23, 2004||Kurd Nasser A.||Method and apparatus for detecting on-die voltage variations|
|US20040268280||Jun 30, 2003||Dec 30, 2004||Eleyan Nadeem N.||On-chip signal state duration measurement and adjustment|
|US20050114056||Dec 31, 2003||May 26, 2005||Patel Jagrut V.||Identifying process and temperature of silicon chips|
|US20050116733||Jan 13, 2005||Jun 2, 2005||Barr Andrew H.||Apparatus and method for detecting and rejecting high impedance failures in chip interconnects|
|US20050174102||Dec 29, 2004||Aug 11, 2005||Stmicroelectronics Pvt. Ltd.||On-chip analysis & computation of transition behaviour of embedded nets in integrated circuits|
|US20050209740||Mar 19, 2004||Sep 22, 2005||Vann Warren E Jr||Systems and methods for controlling fans|
|US20050278520 *||Oct 1, 2004||Dec 15, 2005||Fujitsu Limited||Task scheduling apparatus in distributed processing system|
|US20050289367||Jun 29, 2004||Dec 29, 2005||Clark Lawrence T||System and method for managing power consumption within an integrated circuit|
|US20060066376||Sep 28, 2004||Mar 30, 2006||Narendra Siva G||Frequency management apparatus, systems, and methods|
|US20060149974||Dec 30, 2004||Jul 6, 2006||Efraim Rotem||Device and method for on-die temperature measurement|
|US20060197697||Mar 3, 2006||Sep 7, 2006||Semiconductor Technology Academic Research Center||On-chip signal waveform measurement apparatus for measuring signal waveforms at detection points on IC chip|
|US20060247873||May 2, 2005||Nov 2, 2006||Ati Technologies, Inc.||Apparatus and methods for measurement of analog voltages in an integrated circuit|
|US20070006007||Jun 30, 2005||Jan 4, 2007||Woodbridge Nancy G||Frequency-dependent voltage control in digital logic|
|US20070074216 *||Apr 8, 2005||Mar 29, 2007||Sony Computer Entertainment Inc.||Task management method, task management device, semiconductor integrated circuit, electronic apparatus, and task management system|
|US20070192650 *||May 31, 2006||Aug 16, 2007||Fujitsu Limited||Multi-processing system distributing workload optimally during operation|
|US20070260895 *||May 3, 2006||Nov 8, 2007||Aguilar Maximino Jr||Selection of processor cores for optimal thermal performance|
|US20080004755||Jun 28, 2006||Jan 3, 2008||Dunstan Robert A||Apparatus and method for automatically configuring control of a fan to be exclusively performed by a motherboard|
|US20080136400||Dec 7, 2006||Jun 12, 2008||Taiwan Semiconductor Manufacturing Co., Ltd.||Dynamic voltage and frequency management in integrated circuits|
|CN1716161A||Jun 14, 2005||Jan 4, 2006||英特尔公司||Temperature controller|
|WO2006072106A2||Dec 29, 2005||Jul 6, 2006||Intel Corp||Device and method for on-die temperature measurement|
|1||Gupta and Rincon-Mora, "Predicting the Effects of Error Sources in Bandgap Reference Circuits and Evaluating Their Design Implications" IEEE's Midwest Symposium on Circuits and Systems (MWSCAS), vol. 3, pp. 575-578, Tulsa, Oklahoma, 2002.|
|4||Moore, B.D.; "Tradeoffs in Selecting IC Temperature Sensors", 1999; Elsevier Science; pp. 181-184.|
|5||PCT International Search Report, mailed Feb. 5, 2008, for PCT application EP2008/050919, 3 pages.|
|6||PCT International Search Report, mailed Jun. 16, 2008, for PCT application EP2008/050922, 4 pages.|
|7||Schweber, A/D and D/A Converters: Critical links that just keep getting better, Apr. 1989, 4 pages.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7904702 *||Aug 13, 2008||Mar 8, 2011||Imagination Technologies Limited||Compound instructions in a multi-threaded processor|
|Mar 28, 2014||REMI||Maintenance fee reminder mailed|
|Feb 6, 2007||AS||Assignment|
Owner name: INTERNATIONAL BUSINESS MACHINES CORPORATION, NEW Y
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:SINGH, DEEPAK K.;ATALLAH, FRANCOIS IBRAHIM;REEL/FRAME:018857/0631
Effective date: 20070201